Systems / Scythe

Scythe — onboard vision for kinetic intercept.

Scythe is the Track 1 guidance testbed: a 5" carbon FPV airframe with a Raspberry Pi 5 running YOLO + Δx/Δy at 180+ FPS. Below: the recorded bench demo (hardware) and the live IBVS simulator (simulation). Mapping pixel error to body rates is next—not field-deployed. High-speed aero work runs separately on Sprinter.

Bench demo (hardware) → Live simulator →

Scythe — 5-inch carbon fiber FPV guidance testbed with Raspberry Pi 5 and under-mounted vision camera for bench testing — Scythe · 5" carbon FPV + Pi 5 — bench guidance testbed

Track 1 · Autonomy: Pi 5 · 180+ FPS · bbox + Δx/Δy
Sim proof: IBVS canvas · PX4 / Gazebo SITL
Next: Tethered rates on Scythe · then Sprinter

Hardware bench

Vision-guided interception.

The interceptor uses a vision-based perception and visual-servoing pipeline to detect and track aerial targets. This Track 1 bench / Scythe stack demonstrates real-time YOLO detection and image-plane pixel error (Δx, Δy) at 180+ FPS—the feedback signal for the guidance loop. Detection and error generation only; rates not closed to motors yet.

CV bench · YOLO + image-plane error Prototype · not flight

Computer-vision bench prototype — real-time target detection, tracking, and image-plane error generation for visual servoing.

Simulation view

One control loop, two perspectives.

The same visual-servoing loop, shown from the command picture and the seeker. Left: top-down pursuit geometry. Right: simulated onboard camera driving pixel error toward zero.

Vandaris IBVS · Simulation digital twin Canvas tick @ 30 Hz

TRACKING

ΔX —px

ΔY —px

Error —px

Range —m

Closing —m/s

Lock confidence 0%

Sim runs0 T+0

Browser IBVS simulator · Command picture + seeker view · Shared telemetry bus

Hardware bench: Python / YOLO on Raspberry Pi 5 — bbox + Δx/Δy @ 180+ FPS (not closed on motors yet)
Scythe · Track 1: ARM64 + global shutter · 5" carbon airframe · MAVLink rates next
Track 2 aero: Sprinter (Gen-3) airframe · electronics integrated, pre-flight — Sprinter page →

Last arrow = next milestone · not closed on hardware yet

A · Perception

GPS-denied vision — real-time track and pixel-error from the camera frame (Pi bench demo above; sim IBVS loop below).

B · Guidance

Terminal guidance research for intercept trajectories—aerial and ground-target scenarios—without a human FPV pilot on the final approach.

C · Integration

One software architecture supports multiple scenario classes in simulation—reducing duplicated research paths across air and ground-target studies.

Next system

Sprinter: the high-speed airframe that will carry this guidance stack.

Sprinter airframe → All systems